Anthropic scales Claude Mythos to critical infrastructure in 15+ countries

Anthropic is broadening its Project Glasswing initiative, aimed at identifying and addressing critical software vulnerabilities through artificial intelligence, to include around 150 new organizations across more than 15 countries. This announcement was made on Tuesday, shortly after the company revealed its plans to confidentially file for an initial public offering, following a significant funding round that valued it at nearly $1 trillion. At the core of Project Glasswing is the Claude Mythos model, which Anthropic has branded as its most formidable AI tool to date. This advanced model is capable of detecting thousands of zero-day vulnerabilities within a span of weeks. Earlier this April, 50 initial partners, including U.S. government entities, were granted access to the Claude Mythos Preview, enabling them to scrutinize their codebases for potential security flaws. The latest expansion includes organizations from various sectors such as power, water, healthcare, communications, and hardware—areas that were previously underrepresented in the initiative's first phase. Many of these newly added partners are companies or nonprofits that maintain critical codebases essential for other organizations and government operations. Anthropic emphasized the importance of this initiative, stating, "What each partner has in common is that a successful attack on their codebase could be catastrophic." They estimate that a significant breach could potentially impact over 100 million individuals, raising serious concerns for both global and national security. The expanded group now comprises organizations from U.S. ally nations, including Australia, Canada, France, Germany, Italy, Switzerland, the Netherlands, Spain, Belgium, Sweden, India, Japan, New Zealand, and South Korea. Notable entities that have received access to Mythos include the U.S.-based identity and security management firm Okta, South Korean giants Samsung and SK Hynix, NATO, and the EU’s cybersecurity agency ENISA. In light of this rapid development, Anthropic acknowledges that it anticipates other AI companies will soon introduce similarly capable models. This urgency to fortify Project Glasswing comes as rival OpenAI has launched its own cybersecurity-focused model, GPT-5.5-Cyber, which is currently being tested by a wide range of partners.